Molecular Mechanisms of HipA-Mediated Multidrug Tolerance and Its Neutralization by HipB

See allHide authors and affiliations

Science  16 Jan 2009:
Vol. 323, Issue 5912, pp. 396-401
DOI: 10.1126/science.1163806


Bacterial multidrug tolerance is largely responsible for the inability of antibiotics to eradicate infections and is caused by a small population of dormant bacteria called persisters. HipA is a critical Escherichia coli persistence factor that is normally neutralized by HipB, a transcription repressor, which also regulates hipBA expression. Here, we report multiple structures of HipA and a HipA-HipB-DNA complex. HipA has a eukaryotic serine/threonine kinase–like fold and can phosphorylate the translation factor EF-Tu, suggesting a persistence mechanism via cell stasis. The HipA-HipB-DNA structure reveals the HipB-operator binding mechanism, ∼70° DNA bending, and unexpected HipA-DNA contacts. Dimeric HipB interacts with two HipA molecules to inhibit its kinase activity through sequestration and conformational inactivation. Combined, these studies suggest mechanisms for HipA-mediated persistence and its neutralization by HipB.

Bacteria that are resistant or tolerant to antibiotics are an increasing threat to human health. Indeed, ∼60% of infections in the developed world are caused by biofilms, which exhibit multidrug tolerance (MDT) (1, 2). MDT is caused by the presence of dormant bacterial cells called persisters, which account for only 10–6 to 10–4 cells in a growing population, making MDT difficult to study (35). Persisters are not mutants but phenotypic variants of wild-type cells that evade killing by somehow adopting a transient dormant state (6, 7). Dormancy provides protection because bactericidal antibiotics kill by corrupting their active targets into producing toxic byproducts. These protected persisters can then switch back to the growth phase after the removal of antibiotics, allowing the bacterial population to survive. The first high-persistence allele, hipA7 (high-persistence A), was identified in Escherichia coli and increased the frequency of persistence by 10,000 fold (810). E. coli hipA encodes a 440-residue protein, HipA, which is cotranscribed with a smaller upstream gene, hipB. HipB is an 88-residue protein that represses the hipBA operon by binding cooperatively to four operators upstream of hipBA (11, 12). HipB forms a complex with HipA, and because wild-type HipA cannot be expressed in the absence of HipB because of its deleterious effects on cell growth, hipBA has been categorized as a toxin/antitoxin (TA) module in which HipA, the toxin, is neutralized by the antitoxin, HipB (13, 14). Toxin proteins from chromosomally encoded TA modules, of which more than 10 have been identified in E. coli, appear to promote cell dormancy and may play roles in the development of persistence under certain conditions (5, 7). Chromosomal TA modules can be grouped into three main superfamilies based on whether the toxin has a ribonuclease (RNase)/gyrase–like fold, RNase barnase–like structure, or a PilT N-terminus (PIN) domain (14). The corresponding antitoxins contain DNA binding domains and C termini that are largely unfolded until bound by the toxin (14). HipA and HipB show no homology to any member of these TA superfamilies. Moreover, HipA is one of the few validated biofilm tolerance factors. Indeed, it has been demonstrated that overexpression of the HipA protein leads to MDT in E. coli (2). However, the mechanism of HipA-mediated MDT is unknown.

To delineate the functions of HipA and HipB in MDT, we carried out biochemical and structural studies on HipA and HipA-HipB-DNA complexes. Because of wild-type HipA–mediated persistence, we used the HipA mutant Asp309→Gln309 (D309Q) (referred to as HipA), which can be produced in large quantities in the absence of HipB (15). The structure of HipA was solved to 1.54 Å resolution and refined to an Rwork/Rfree of 19.5/23.2% (table S1 and fig. S1) (1618). The HipA structure has a globular fold with 15 β strands and 15 α helices and can be divided into an N-terminal α/β domain and an all–α-helical C-terminal domain (Fig. 1A). Density is missing for residues 185 to 195, which are near the active site and probably correspond to the activation loop of other kinases. Structure-based homology searches revealed that HipA is most similar to human CDK2/cyclin A kinase (19). The structural homology between HipA and CDK2 was highest in the C-terminal region that contains CDK2 catalytic residues, suggesting that HipA functions as a protein kinase, as reported (15). Although HipA is most similar to CDK2, the proteins superimposed with a large root mean square deviation (RMSD) of 3.9 Å for 150 corresponding Cα atoms, indicating that HipA represents a previously unknown class of protein kinase (20).

Fig. 1.

HipA is a protein kinase that phosphorylates EF-Tu. (A) The E. coli apoHipA structure. β sheets and α helices are colored magenta and cyan, respectively. Secondary structural elements and the N and C termini are labeled. Disordered loops, including the putative activation loop (labeled AL), are indicated by dashed lines. (A) to (C) and Figs. 2, A to C; 3, B to D; and 4, A to D were made using PyMOL (31). (B) The ATP-binding pocket of HipA. Shown are ATP molecules (sticks), Mg2+ ions (magenta spheres), water molecules (red spheres), and hydrogen bonds (dashed black lines). (C) Superimposition of the C domains of apoHipA (yellow) and HipA-ATP-Mg2+ (blue) reveals only a small rotation upon ATP binding. ATP in the HipA-ATP structure is shown as sticks. (D) (Top) HipA–(GST-EF-Tu) pulldown (SDS-PAGE, stained with Coomassie brilliant blue). Lanes are as follows: 1, molecular weight ladder; 2, glutathione agarose (bead) retentate after addition of GST-EF-Tu, HipA, ATP, and GDP; 3, bead retentate after addition of GST-EF-Tu, HipA, ATP, and GDP, and washing; 4, bead retentate after addition of GST-EF-Tu, HipA, and no GDP or ATP; 5, bead retentate after addition of GST-EF-Tu, HipA, and no GDP or ATP, and washing. (Bottom) Wild-type HipA kinase assay using EF-Tu as a substrate (immunoblot). The positions of GST-EF-Tu and Xa-cleaved EF-Tu are indicated by red and blue arrows, respectively. Lanes are as follows: 1, EF-Tu (cleaved) + wild-type HipA + ATP + GTP; 2, EF-Tu (cleaved) + wild-type HipA + ATP + GDP; 3, EF-Tu (cleaved) + inactive HipA (D309Q) + ATP + GDP; 4, EF-Tu (uncleaved) + wild-type HipA + ATP + GDP; 5, EF-Tu (uncleaved) + wild-type HipA + ATP + GTP; 6, wild-type HipA + ATP; 7, native EF-Tu (cleaved) + ATP + GDP.

HipA contains all the catalytic residues found in protein kinases, including the putative catalytic base Asp309 (20). The D309Q mutation abrogates persistence, strongly suggesting that kinase function is key to HipA-mediated MDT (15). Indeed, we found that HipA binds adenosine triphosphate (ATP) with a dissociation constant (Kd) of 18.0 ± 2.0 μM, which is similar to the dissociation constants obtained for ATP binding to other serine/threonine kinases (fig. S2) (20). To delineate the ATP binding mechanism of HipA, we determined the structure of the HipA-ATP-Mg2+ complex to 1.66 Å resolution and refined the structure to an Rwork/Rfree of 18.4/21.7%. Density for ATP is observed in the cleft between the HipA N and C domains (fig. S1). HipA binds ATP with high selectivity (Fig. 1B). Specifying contacts are provided to the adenine N6, N1, and N3 atoms by the carbonyl oxygen of Glu234, the amide nitrogen of Phe236, and the side chain Nϵ of Gln252, respectively. The adenine ring stacks with Phe236 and Tyr331, whereas Val98, Val151, Ile179, and the side-chain methylene carbons of salt-bridged residues Asp237 and Arg235 provide hydrophobic interactions. The γ phosphate hydrogen-bonds to both the side chain of His311 and the amide groups of Gly153 and Ala154, which form part of a loop analogous to the Gly loops of other protein kinases (Fig. 1B). Residues 152 to 156 of this loop are less ordered in the substrate-free HipA (apoHipA) structure, indicating that nucleotide binding is required for its stabilization. Two Mg2+ ions are also present in the HipA-ATP structure and probably function analogously to other protein kinases in facilitating phosphotransfer by accelerating substrate association and product dissociation (20, 21).

Comparison of the HipA-ATP and apoHipA structures revealed that binding ATP causes the N and C domains to undergo only a small rotation (∼4°) relative to each other (Fig. 1C). However, by analogy to other kinases, a more pronounced closure of the HipA domains upon binding the protein substrate is expected (20). The findings of a specialized kinase fold and high-affinity ATP binding strongly supported the hypothesis that HipA mediates persistence by phosphorylating one or more target proteins. To identify possible HipA targets, we carried out in vitro pulldown assays on candidate E. coli proteins. One protein, EF-Tu, was found to interact strongly with HipA in the presence of ATP-Mg2+ and guanosine diphosphate (GDP) (Fig. 1D). EF-Tu, the most abundant protein in E. coli, belongs to the guanosine triphosphatase superfamily and plays an essential role in translation by catalyzing aminoacyl–transfer RNA (-tRNA) binding to the ribosome (22). Upon guanosine triphosphate (GTP) hydrolysis to GDP, EF-Tu undergoes a conformational change to an open form, which cannot bind the ribosome. Previous studies showed that EF-Tu is phosphorylated on residue Thr382 by an unknown kinase or kinases (23, 24). The side chain of Thr382 contacts Glu117 to stabilize the GTP-bound closed state of EF-Tu. Phosphorylation of Thr382 favors the GDP-bound open form because it would lead to repulsion of Glu117 and prevent EF-Tu from adopting the GTP-bound closed conformation. Thr382-phosphorylated EF-Tu cannot bind aminoacyl-tRNA and is therefore inactive in translation (23, 24). To test whether EF-Tu is a HipA substrate, we used an in vitro transcription/translation system to produce the toxic wild-type HipA enzyme (fig. S3). Immunoblotting studies, using antibodies to pThr/pSer/pTyr, indicated that HipA could phosphorylate EF-Tu in a manner stimulated by GDP (Fig. 1D). Moreover, fluorescence polarization studies revealed that HipA bound the EF-Tu peptide, IREGGRTVGA (25), encompassing Thr382 (shown in bold here) with a Kd of 15 ± 5 μM (fig. S4). Subsequently, we solved a crystal structure of the HipA–(AMP-PNP)–IREGGRTVGA complex to 3.5 Å resolution. The structure revealed that the activation loop, residues 185 to 195, was now folded and density was observed for the peptide near the active site and close to the activation loop (fig. S4). These combined data suggest that HipA may phosphorylate Thr382 to block aminoacyl-tRNA binding by EF-Tu. However, given that HipA affects multiple E. coli processes, other cellular targets are likely (9, 10).

Under normal cellular conditions, the persistence function of HipA is somehow masked by its tight interaction with HipB (11, 12). HipB also functions as a transcriptional autoregulator of the hipBA operon by cooperatively binding four operators with the consensus sequence TATCCN8GGATA (where N indicates any nucleotide), located in the hipBA promoter region (11, 12). HipB binds these operators with high affinity, which is enhanced by the addition of HipA to the complex (12). To delineate the mechanism of HipB-mediated inhibition of HipA, the structure of the HipA-HipB complex bound to a 21–base pair hipB operator (top strand ACTATCCCCTTAAGGGGATAG) was solved and refined to an Rwork/Rfree of 22.5/28.1% to 2.68 Å resolution (table S1) (Figs. 2 and 3).

Fig. 2.

Crystal structure of the HipA-HipB-DNA complex. (A) Ribbon diagram of the HipA-HipB-DNA operator complex. The two HipA monomers are blue, and one subunit of the HipB dimer is yellow and the other orange. The N and C termini and secondary structural elements of one HipB subunit (orange) are labeled. β1′ is labeled for the yellow subunit. Also labeled are the N and C domains of each HipA molecule. The DNA is shown as sticks with carbon, nitrogen, oxygen, and phosphorus atoms colored green, blue, red, and magenta, respectively. (B) Superimposition of substrate-free HipA (red) onto HipB-bound HipA (blue), showing their essentially identical conformations. For clarity, only one HipA molecule in the HipA-HipB-DNA complex is shown. (C) Closeup of the HipA-HipB interaction interface. Each lateral side of the HipB dimer, labeled subunits 1 and 2, interacts with the N and C domains of one HipA monomer. The DNA is shown as a gray surface. For clarity, only one HipA molecule is shown because the interaction interface between the other HipA molecule and the HipB dimer is identical. The residues that contribute to the interface are labeled and shown as orange sticks for HipB and dark blue sticks for HipA.

Fig. 3.

HipB and HipA interactions with the hipB operator DNA. (A) Schematic representation of HipB-HipA-DNA interactions. Only one half site of the 21-oligomer duplex is shown because the identical contacts are made with each half site. The strands are labeled 1A to 10A and 1B to 10B. Bases are represented as rectangles and labeled according to sequence. The ribose groups are shown as pentagons. The operator signature motif sequence, TATCC, is red. HipB-DNA contacts are yellow. Hydrophobic contacts are indicated by lines and hydrogen bonds are indicated by arrows. Blue arrows indicate HipA-phosphate contacts. (B) HipB-DNA interactions. Only one HipB subunit-DNA half site is shown. The DNA and residues making side-chain contacts are shown as sticks. The signature motif sequence, TATCC, is labeled in red. For clarity, only the four-helix bundle is shown and labeled. (C) HipA-DNA contacts. The HipB dimer (yellow) is shown for reference. The location of the two DNA-interacting residues from HipA, K379 and R382, are shown as blue sticks and highlighted by mesh surface representations. The DNA is shown as sticks and colored as in (B). (D) HipA-HipB bound DNA is bent. This is an omit Fo-Fc map in which the DNA was omitted from refinement to 2.68 Å resolution. The map is contoured at 2.8 σ. The DNA is shown as sticks and the bend angle is indicated.

HipB forms a compact dimer that specifically interacts with DNA through major groove contacts, whereas two HipA molecules sandwich the HipB-DNA complex by contacting the sides of the HipB dimer (Fig. 2). HipB binds far from the HipA active sites and, unlike other TA inhibition mechanisms, does not occlude the active site. The HipB dimer interface is extensive and buries 2700 Å2 of accessible surface area (ASA), which accounts for over 36% of the total dimer ASA. HipB contains one β strand and four α helices with topology α1–α2–α3–α4–β1. Helices 2 and 3 form a canonical helix-turn-helix (HTH) motif. The first 3 and last 16 residues of each HipB subunit are disordered and located near a small β sheet that is composed of β1 and β1′ (from the other subunit) and forms a “β lid” (Fig. 2A). The HipB subunit structure showed significant homology to 434 Repressor, 434 Cro, and the restriction-modification controller protein C. AhdI from Aeromonas hydrophila, with RMSDs of 1.56, 1.60, and 1.51 Å for 59, 56, and 59 corresponding Cα atoms, respectively, thus placing it in the Xre-HTH family of transcriptional regulators (26). The homology between HipB and these proteins is confined to the four-helix bundle region because the β lid is found only in HipB. Despite the similarities in DNA binding domains, these proteins bind their DNA sites differently because 434 Cro does not significantly distort its DNA site, and biochemical data indicate that C. AhdI bends its DNA site by 47° (27, 28). In contrast, HipB induces a large, 70° bend in its operator (Fig. 3D). This bending may play a role in the cooperative binding of HipB to its four operator sites, which is predicted to involve DNA wrapping (11, 12). Indeed, the hipBA promoter also contains a binding site for the architectural protein IHF, which could further aid in DNA condensation.

HipB-induced DNA distortion aligns the recognition helices for specific binding to consecutive major grooves. Contacts from the HTH motif completely specify the nucleotides of the HipB signature motif, T2A3T4C5C6 (Fig. 3, A and B). Ser29 from α2 makes hydrophobic contacts with the Thy2 methyl group. Residues of the recognition helix provide the remaining base-specifying contacts whereby two hydrogen bonds from Gln39 read Ade3, whereas two hydrophobic contacts from Ala40 and Ser43 specify Thy4. Finally, Lys38 makes hydrogen bonds with the guanine O6 oxygens of base pairs 5 and 6. HipB also makes 11 phosphate contacts to each half site. Deoxy-ribonuclease I protection studies showed that HipA binding to the HipB-DNA complex leads to an increase in protection and binding affinity (12). This is explained by the finding that HipA provides four phosphate backbone contacts to each half site from Lys379 and Arg382 (Fig. 3, A and C).

In the HipA-HipB-DNA complex, the HipB dimer is sandwiched on each side by one HipA molecule, and the complex is formed from noncontiguous regions of both HipA and HipB (Figs. 2 and 3C). This type of interaction contrasts sharply with structures of other TA modules in which the toxin interacts with a C-terminal region of the antitoxin that typically is structured only in the presence of toxin. Specifically, for the HipA-HipB pair, the HipA N domains interact with one HipB subunit, whereas the HipA C domains interact primarily with the other HipB subunit (Fig. 2C). This interaction interface is extensive, burying ∼5000 Å2 of ASA, and involves both nonpolar and polar interactions. In the HipA N-domain–HipB interface, HipB residues from the turn before α1 interact with residues on HipA β15, and residues on HipB α1 make extensive contacts with HipA residues located on a long 310-like loop between β3 and α1. The formation of the HipA C-domain–HipB interface primarily involves HipB residues from α2 and the turn between α4 and β1. These residues interact primarily with HipA residues in the loop between α8 and α9 and the N terminus of α9. In addition, “cross-subunit” contacts are made between HipB residues Gln12 and HipA C-domain residue Gly284, and between HipB residue Tyr8 and HipA C-domain residue Ser286. These cross contacts, combined with the numerous interactions of each subunit in the HipB dimer with the N and C domains of HipA, lock HipA in an open and probably inactive conformation (Fig. 2C).

To activate HipA for persistence and free it from its DNA tether, HipB must be removed or degraded. Unlike most antitoxins, HipB interacts with HipA using residues from noncontiguous, well-ordered domains and not loops. Proteases that degrade toxins typically bind and tug on disordered regions to unfold the substrate. HipB contains an exposed and flexible 16-residue C terminus attached to the small β lid that covers the hydrophobic core of the protein and would appear to be an excellent candidate for protease attack (Fig. 4A). The structure of the HipA-HipB-DNA complex also provides insight into the mechanism of increased persistence of the HipA7 protein. HipA7, which contains two substitutions, G22S and D291A, confers a high-persistence phenotype on E. coli cells independent of HipB. Subsequent data revealed that the D291A mutation alone was sufficient for this phenotype (29). The HipA-HipB-DNA structure indicates that this phenotype probably results from a weakened HipA-HipB interaction, which unleashes HipA kinase activity. Specifically, Asp291 makes key contacts to stabilize the HipA-HipB interface, including hydrogen bonds to the side chain of Ser285, which positions the Ser285 carbonyl oxygen to interact with HipB residue Gln62, and hydrogen bonds to the amide nitrogen of Leu327, which buttresses the HipA C-terminal region that interacts with HipB (Fig. 4B).

Fig. 4.

HipB is a vulnerable antitoxin that neutralizes HipA. (A) The HipB dimer is shown as ribbons with one yellow and one orange subunit. The DNA is shown as pink sticks. Hydrophobic core residues are shown as sticks and transparent surfaces. The β lid is composed of a short two-stranded β sheet. Disordered C-terminal residues extend from the lid and are depicted as colored dashes. (B) Mutation sites (G22S and D291A) in the HipA7 protein mapped onto HipA proteins of the HipA-HipB-DNA complex. The HipA molecules are cyan, the HipB subunits are yellow and orange, and the DNA is a pink ribbon. Residues G22 and D291 are shown as blue Corey, Pauling, Koltan structures (CPKs). HipB residue Q62 and HipA residues S285 and L327 are also shown as CPKs. A closeup of the interactions involving HipA residue D291 is shown (inset). Hydrogen bonds are indicated by dashed lines. (C) Structure of the HipA(ATP)-HipB-DNA complex. HipA molecules are yellow, the HipB dimer is cyan, and the DNA and ATP are shown as sticks. The blue mesh represents a 2.98 Å resolution Fo-Fc map, contoured at 4.0 σ, in which the ATP was omitted. (D) Closeup view of the ATP-binding pocket and omit map. Carbon, nitrogen, oxygen, and phosphorus atoms are colored magenta, blue, red, and purple, respectively.

HipA undergoes only a small conformational change upon binding ATP, suggesting that HipA could bindATP when in complex with HipB-DNA. Indeed, HipA(ATP)-HipB-DNA crystals isomorphous to HipA-HipB-DNA crystals could be grown de novo. Alternatively, ATP could be soaked into preformed HipA-HipB-DNA crystals. In both cases, difference Fo-Fc electron density maps revealed clear density for ATP in the HipA active site (Fig. 4, C and D). In addition, isothermal titration calorimetry studies revealed a Kd of 15.0 ± 1.0 μM for ATP-Mg2+ binding to HipA in the HipA-HipB-DNA complex, which is essentially identical to that obtained for ATP-Mg2+ binding to HipA alone (fig. S2). If the HipA active site is not blocked for ATP binding, then how does HipB binding neutralize HipA? Data from other protein kinase structures indicate that although ATP binding causes only small domain movements such as those we observed in HipA, the binding of protein substrates causes substantial domain closure (20, 21). This large-scale movement brings the two substrates into proximity for catalysis and precludes bulk solvent from the active site. HipB binding would appear to prevent this conformational change in HipA by locking the enzyme into an inactive open conformation by its extensive interactions with the HipA N and C domains. Finally, the recent finding that E. coli EF-Tu is localized primarily to the cytosolic and membrane fractions, which is far from the nucleoid where the HipA-HipB-DNA complex would reside, suggests that HipB-DNA binding may also inactivate HipA by its sequestration (30).

These studies have provided important insight into the mechanisms by which HipA mediates persistence and HipB neutralizes HipA. The high conservation of HipA among Gram-negative bacteria indicates its central role in the development of persistence. Thus, inhibitors that specifically target the substrate-binding sites of HipA, may prove effective against persistence and MDT.

Supporting Online Material

Materials and Methods

Figs. S1 to S4

Table S1

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article